Method, system, equipment and medium for high-throughput accurate quantitative identification of poplar property

Abstract

The invention relates to a poplar property high-throughput accurate quantitative identification method, system, equipment and medium, the innovation of the method lies in that spectral data analysis and automatic script processing are combined, efficient data acquisition, import, matching and storage are realized, abnormal point detection is carried out by introducing principal component analysis and anti-noise mahalanobis distance, a data set is optimized, and the identification accuracy of the poplar property is improved. Building and training a model to realize high-flux accurate prediction of the material property and the density; a sample control module, an acquisition test module, an automatic processing module, a data set determination module and a result output module are integrated on the system, so that full-process automatic and intelligent management is realized; the equipment comprises a database and a memory, and the method is cooperatively executed through the database and the memory to ensure efficient operation of the method; a computer readable medium stores executable instructions, which is convenient for wide deployment and application. The method is excellent in high-throughput analysis, cost saving, environmental protection and data accuracy, and development of the poplar property identification technology is effectively promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of wood identification, and in particular to a method, system, equipment and medium for high-throughput, precise and quantitative identification of poplar wood properties. Background Art

[0002] At present, poplar, as the main source of timber for my country's artificial forests, is widely favored for its tall trunks, luxuriant branches and leaves, strong adaptability, rich genetic diversity, and especially its fast-growing characteristics. Poplars not only have extremely high economic value, but also play an important role in ecology and society. Among many applications, plywood and pulping and papermaking are the main uses of poplars. It is worth mentioning that the lignin content in plywood has a significant effect on its gluing properties, while hemicellulose plays a role in enhancing the bonding strength of paper during the pulping process. Therefore, it is particularly important to quickly and quantitatively identify these chemical components in the raw materials.

[0003] In addition, the density of wood is a key indicator for measuring its mechanical properties and quality. Its accurate prediction and evaluation has far-reaching significance for improving wood materials, guiding artificial afforestation, improving wood utilization efficiency and extending its service life. Therefore, in modern forestry production, it has become an urgent need to be able to quickly and accurately obtain wood material information.

[0004] However, traditional quantitative methods for wood properties have many shortcomings. These methods are not only complex, time-consuming and labor-intensive, but also require the consumption of large amounts of chemicals and produce large amounts of organic waste liquid, posing a threat to the environment. In addition, these methods also require professional laboratory personnel to operate chemical equipment, further increasing the limitations of their application. Therefore, such methods are only applicable to dedicated laboratory environments for small samples, cannot meet the needs of real-time analysis, and are even more difficult to achieve rapid and accurate analysis of a large number of samples. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a high-throughput and accurate quantitative identification method, system, equipment and medium for poplar wood properties, which solves the technical problem that traditional wood property quantitative methods have the problems of complex operation, time-consuming and labor-intensive, large consumption of chemicals, serious environmental pollution and high application limitations, and are difficult to meet the needs of modern forestry for rapid and accurate acquisition of wood property information.

[0006] (II) Technical solution In order to achieve the above object, the main technical solutions adopted by the present invention include: In a first aspect, an embodiment of the present invention provides a method for high-throughput quantitative identification of poplar wood properties, comprising: automatically identifying and selecting poplar log segments that meet preset conditions, and controlling processing equipment to process the poplar log segments into samples; automatically searching for and marking areas to be collected that meet preset conditions on at least a portion of the samples, directly collecting spectral data of unlabeled samples or collecting spectral data of labeled samples in the area to be collected through a spectrometer, and performing density tests and / or multiple batches of wood property tests on the samples; using a pre-built automated script to automatically import, match, and store spectral data, density data, and chemical composition data; introducing principal component analysis and anti-noise Mahalanobis distance to detect and remove outliers from the stored data, and then obtaining a data set after preprocessing and spectral wavelength selection; based on the data set, constructing and training a hybrid intelligent learning model based on stacked autoencoders and random forests for the spectral data and test data of poplars, so as to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplars.

[0007] Optionally, automatically identifying and selecting poplar log segments that meet preset conditions, and controlling processing equipment to process the poplar log segments into samples includes: using a sensor array to automatically identify and select poplar log segments that meet preset conditions; sending instructions to the processing equipment based on the identification results, and controlling the processing equipment to process the selected poplar log segments into blank samples of predetermined sizes according to a preset processing flow; after the blank samples are air-dried, controlling the processing equipment to process the blank samples into a number of wood blocks, and automatically dividing the several small wood blocks into a first part of the sample and a second part of the sample, and uniquely marking each part of the sample; controlling the grinder to grind the second part of the sample into wood powder, and controlling the filtering equipment to filter the wood powder into a sieve with a pre-specified mesh number to obtain filtered wood powder.

[0008] Optionally, automatically searching and marking the area to be collected that meets the preset conditions on at least a part of the samples, directly collecting the spectral data of the unmarked samples or collecting the spectral data of the area to be collected of the marked samples through the spectrometer, and performing density test and / or material property test on the samples in multiple batches, including: after the first part of the samples are placed in a constant temperature and humidity chamber for equilibrium treatment, controlling the density detection equipment to perform density standard value detection on the first part of the samples, and recording the obtained density data as a reference label of the model; starting and controlling the spectrometer to preheat for 25 to 35 minutes, and after the preheating is completed, calibrating the light source and instrument of the spectrometer through a polytetrafluoroethylene reference white board; controlling the spectrometer in diffuse reflection mode, and performing near-infrared spectrum collection on the filtered wood powder in a constant temperature chamber; The surface information of the first part of the sample is read by a camera deployed at the optical fiber probe of the spectrometer, and defects, blue discoloration, heartwood or sapwood are identified again to search for the area to be collected; the spectrometer is controlled to collect near-infrared spectral data of the cross section of the first part of the sample in the area to be collected; and the material properties of multiple batches of filtered wood powder are tested according to preset standards, wherein the material properties of the multiple batches include the content of lignin, cellulose and hemicellulose.

[0009] Optionally, using a pre-built automation script to automatically import, match and store spectral data, density data and chemical composition data includes: starting a pre-configured Python script by timed or event triggering, and automatically detecting and reading new result files generated by a spectrometer, a material property tester and a density detection device by using an incremental import mechanism; using an asynchronous coroutine framework of the Python script to create an asynchronous task queue, adding each new result file detected as an independent task to the asynchronous task queue in parallel, and in an asynchronous loop, continuously taking out new result files from the asynchronous task queue and importing data from the new result files; using a Python script to create and initialize an empty dictionary, configuring the unique tag of the sample as the key of the dictionary, and initializing the value of the dictionary to an empty list for storing various types of imported data associated with the key; traversing the unique serial number of each newly detected sample, and checking whether the unique serial number of each sample already exists in the dictionary: if the unique serial number does not exist in the dictionary, creating a new serial number entry for the unique serial number, and adding the imported spectral data, chemical composition data and density data to the list corresponding to the new serial number entry in parallel; If the unique serial number already exists in the dictionary, choose to skip the serial number entry, append a new serial number entry, or update the list data corresponding to the serial number entry; use Python scripts to scan and verify data integrity, and check whether the data under each list is complete through pre-defined integrity rules; the pre-defined integrity rules include non-empty field check, data range verification, and format verification; for incomplete data detected, add a pre-specified label or move to a new temporary data table for storing incomplete data; use Python scripts to perform duplicate sample detection, automatically identify and mark duplicate data with unique serial numbers by adding another pre-specified label, and prompt the user to check or perform a merge operation; after the verification is completed, use Python scripts to save the verified spectral data, density data, and chemical composition data to the specified output file.

[0010] Optionally, principal component analysis and anti-noise Mahalanobis distance are introduced to detect and remove abnormal points from the stored data, including: constructing a spectral data matrix based on the stored spectral data, and calculating the covariance matrix of the spectral data matrix; performing eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues ​​and corresponding eigenvectors; selecting eigenvectors that can explain a predetermined proportion of the total variance according to the size of the eigenvalues ​​to form a transformation matrix; projecting the spectral data matrix into a low-dimensional space composed of the selected eigenvectors according to the transformation matrix to obtain a score matrix after dimensionality reduction; in the principal component space, introducing the anti-noise Mahalanobis distance algorithm, and using a method based on the median and absolute median difference to adjust the mean and covariance matrix of the score matrix; for each spectral data, calculating the Mahalanobis distance based on the mean and covariance matrix of the score matrix obtained based on the median and absolute median difference; determining a basic threshold based on the mean and standard deviation of the Mahalanobis distance; determining the position of each spectral data in the principal component space; k Nearest neighbors, calculate each spectral data to the corresponding k The distances of the nearest neighbors are calculated and averaged to obtain local aggregation information; The basic threshold is dynamically adjusted according to the local aggregation information to obtain the current dynamic threshold; if the Mahalanobis distance of the sample exceeds the current dynamic threshold, it is marked as an abnormal point, and all spectral data marked as abnormal are deleted from the output file storing the spectral data to obtain the cleaned spectral data; in, The spectral data matrix is: ; In the formula, X is the spectral data matrix, x ij Indicates i The spectral data samples are j The value of a spectral feature; The covariance matrix of the spectral data matrix is: ; In the formula, ∑ is the covariance matrix, X is the spectral data matrix, X T is the transpose of matrix X, n is the number of samples; The characteristic values ​​are: ; Eigenvectors: ; Number of eigenvectors k The conditions are: ; In the formula,p Spectral feature number; The score matrix after dimensionality reduction is: ; In the formula, S is the score matrix after dimensionality reduction, X is the spectral data matrix, E is the transformation matrix composed of eigenvectors; The Mahalanobis distance is: ; In the formula, D i is a sample of spectral data i The Mahalanobis distance, S i is a sample of spectral data i The score vector of and are the mean and covariance matrices of the score matrix, respectively. is the inverse matrix of the covariance matrix; The basic threshold is: ; In the formula, C is the dynamic threshold, mean is the mean Mahalanobis distance of all samples, std is the standard deviation of the Mahalanobis distance of all samples; The local aggregation information is: ; In the formula, ρ i is the spectral data sample point i The local aggregation information of KNN (i) is the spectral data sample point i of k The set of nearest neighbors, d ( i , j ) is the spectral data sample point i and j The distance between The dynamic thresholds are: ; In the formula, is the spectral data sample point i Dynamic anomaly detection threshold, is an adjustment factor greater than 0.

[0011] Optionally, after applying preprocessing and spectral wavelength selection, the data set is obtained, including: performing one or more preprocessing including noise removal, smoothing, baseline correction, standard variable transformation, and standard normal transformation on the cleaned spectral data; calculating the information gain value between each wavelength in the preprocessed spectral data and the target variable one by one; sorting all wavelengths according to the calculated information gain value, and screening the wavelengths ranked in the top γ as characteristic wavelengths, where γ is a sorting threshold set according to demand or expert experience; using a similarity measure based on spectral intensity values, calculating the correlation between the remaining wavelengths that have not been selected as characteristic wavelengths and the target variable; sorting the remaining wavelengths according to the size of the correlation, and screening the wavelengths ranked in the top λ The wavelengths are taken as the wavelengths that are viscously related to the target variable; a redundancy check is performed on the wavelengths that are viscous related to the target variable to exclude wavelengths whose differences with the characteristic wavelengths are less than or equal to the set difference threshold or are repeated with the characteristic wavelengths; after the redundancy check, the spectral data are reconstructed according to the remaining non-redundant wavelengths that are viscous related to the target variable and the selected characteristic wavelengths; the training set and the test set are obtained according to the reconstructed spectral data, density data and chemical composition data.

[0012] Optionally, according to the data set, constructing and training a hybrid intelligent learning model based on stacked autoencoders and random forests for the spectral data of poplar and the test data to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplar includes: obtaining training set data containing spectral data and chemical composition data labels from the data set, and dividing to obtain a first feature matrix of the spectral data X chem and the label vector of chemical composition data Y chem ; Obtain training set data containing reference labels of spectral data and density data from the data set, and divide it to obtain the second feature matrix of spectral data X density and the label vector of the density data Y density ; Determine and construct two parallel autoencoder paths of the stacked autoencoder, each autoencoder path includes multiple network layers for extracting data features layer by layer; randomly initialize the weights and biases of each autoencoder path; convert the first feature matrix X chem and the second characteristic matrix X density Input into two parallel autoencoder paths respectively, train them independently, and optimize the network parameters of the two autoencoder paths by minimizing the reconstruction error of each path; Apply the trained stacked autoencoder to the first feature matrix X chemand the second characteristic matrix X density , respectively extract the compressed first deep feature representation F chem And the compressed second deep feature representation F density ; Use the extracted first feature representation F chem and the label vector of chemical composition data Y chem , train a random forest regression model for chemical composition prediction, and use the extracted second feature representation F density and the label vector of the density data y density , train a random forest regression model for density prediction; use the cross-validation method to evaluate the two random forest regression models, and optimize the model performance by adjusting the model parameters; for the new poplar spectral data input, first extract features through the trained stacked autoencoder, and then input the extracted new features into the optimized random forest regression model for prediction, and output the quantitative prediction results of multiple batches of chemical composition data and density data respectively; the formula for minimizing the reconstruction error is: ; In the formula, α is the weight parameter, N is the number of data points of the original data, is the original input data, is the reconstructed data output by the autoencoder, is the mean of the original data, is the mean of the reconstructed data, is the variance of the original data, is the variance of the reconstructed data, is the covariance between the original and reconstructed data, c 1 and c 2 is a constant term introduced to avoid division by zero errors.

[0013] In a second aspect, an embodiment of the present invention provides a high-throughput quantitative identification system for poplar wood properties, which is applied to implement the high-throughput quantitative identification method for poplar wood properties as described above, including: a sample control module, which is used to automatically identify and select poplar log segments that meet preset conditions, and control processing equipment to process the poplar log segments into samples; a collection and testing module, which is used to collect spectral data of samples through a spectrometer, and perform density tests on a part of the samples, and perform multiple batches of wood property tests on another part of the samples; an automation processing module, which is used to automatically import, match and store spectral data and chemical composition data using a pre-built automation script; a data set determination module, which is used to introduce principal component analysis and anti-noise Mahalanobis distance, detect and remove outliers on the stored data, and then obtain a data set after preprocessing and spectral wavelength selection; a result output module, which is used to construct and train a hybrid intelligent learning model based on stacked autoencoders and random forests for the spectral data and test data of poplar according to the data set, so as to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplar.

[0014] In a third aspect, an embodiment of the present invention provides a high-throughput quantitative identification device for poplar wood properties, comprising: a multimodal database cluster, including at least one database, the multimodal database cluster supporting distributed storage and parallel processing; and an elastic cloud storage communicatively connected to the multimodal database cluster; wherein the elastic cloud storage stores instructions that can be executed by at least one database, and the instructions are executed by at least one database so that the at least one database can execute the high-throughput quantitative identification method for poplar wood properties as described above.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer-readable medium having computer-executable instructions stored thereon, and when the executable instructions are executed by a processor, the high-throughput quantitative identification method for poplar wood properties as described above is implemented.

[0016] (III) Beneficial effects The beneficial effects of the present invention are as follows: first, it can automatically identify and select poplar logs that meet preset conditions, and process them into standardized samples through processing equipment, thereby ensuring the consistency and accuracy of the analysis. Second, the spectrometer can quickly collect spectral data of the sample, realizing non-destructive rapid detection and greatly improving the detection efficiency.

[0017] Secondly, the spectrometer is used to quickly collect spectral data of samples, and density tests are performed, as well as multiple batches of material property tests are performed in combination with preset standards, so that the present invention can efficiently and batch-wise obtain material property information and density data of poplars.

[0018] In addition, the use of pre-built automation scripts to automatically import, match and store spectral data, density data and chemical composition data further improves the degree of automation in data processing, while reducing dependence on the professional level of experimenters and reducing the complexity and error rate of manual operations.

[0019] More importantly, the accuracy and reliability of the data are ensured by introducing principal component analysis and noise-resistant Mahalanobis distance for outlier detection and removal. The dataset obtained after preprocessing and spectral wavelength selection provides strong support for building an accurate model.

[0020] Furthermore, the present invention provides an innovative hybrid intelligent learning model that integrates stacked autoencoders and random forest algorithms. This model can accurately predict the complex relationship between multiple batches of chemical composition data and density data and spectral data of poplars. Through this technological breakthrough, we have achieved high-throughput analysis of a large number of poplar samples, which not only significantly improves the detection efficiency, but also greatly reduces the detection cost, and reduces the dependence of experiments on chemicals, meeting environmental protection requirements. It realizes the quantitative prediction of the relationship between large batches of poplar wood properties and spectral data, greatly improves the adaptability and prediction accuracy of the model, ensures the reliability and stability of the results, simplifies the operation process, and reduces the professional requirements for experimenters.

[0021] Therefore, the high-throughput quantitative identification method of poplar wood properties provided by the present invention shows significant beneficial effects in high-throughput rapid analysis, cost reduction, environmental protection, data accuracy and model prediction accuracy, providing important technical support for forest breeding and resource optimization management, and effectively solving the problems existing in traditional wood property quantitative methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a flow chart of a method proposed in an embodiment of the present invention; Figure 2 A specific flow chart of step S1 of the method proposed in an embodiment of the present invention; Figure 3 A schematic diagram of a specific flow chart of step S2 of the method proposed in an embodiment of the present invention; Figure 4 A schematic diagram of a specific flow chart of step S3 of the method proposed in an embodiment of the present invention; Figure 5 A schematic diagram of a specific flow chart of step S4 of the method proposed in an embodiment of the present invention; Figure 6 This is a schematic diagram of a specific flow chart after step S4 of the method proposed in an embodiment of the present invention; Figure 7 A schematic diagram of a specific flow chart of step S5 of the method proposed in an embodiment of the present invention; Figure 8 This is a comparison chart of the predicted value and the actual value of lignin according to the method proposed in the embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0024] like Figure 1 As shown, a high-throughput quantitative identification method for poplar wood properties proposed in an embodiment of the present invention includes: automatically identifying and selecting poplar log segments that meet preset conditions, and controlling processing equipment to process the poplar log segments into samples; automatically searching and marking areas to be collected that meet preset conditions on at least a portion of the samples, directly collecting spectral data of unlabeled samples or collecting spectral data of labeled samples in the area to be collected through a spectrometer, and performing density tests and / or multiple batches of wood property tests on the samples; using pre-built automated scripts to automatically import, match and store spectral data, density data and chemical composition data; introducing principal component analysis and anti-noise Mahalanobis distance to detect and remove outliers from the stored data, and then obtaining a data set after preprocessing and spectral wavelength selection; according to the data set, constructing and training a hybrid intelligent learning model based on stacked autoencoders and random forests for the spectral data of poplars and the data obtained from the tests, so as to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplars.

[0025] First, it can automatically identify and select poplar logs that meet the preset conditions, and process them into standardized samples through processing equipment, thereby ensuring the consistency and accuracy of the analysis. Second, the spectrometer can quickly collect the spectral data of the samples, realizing non-destructive rapid detection and greatly improving the detection efficiency.

[0026] Secondly, the spectrometer is used to quickly collect spectral data of samples, and density tests are performed, as well as multiple batches of material property tests are performed in combination with preset standards, so that the present invention can efficiently and batch-wise obtain material property information and density data of poplars.

[0027] In addition, the use of pre-built automation scripts to automatically import, match and store spectral data, density data and chemical composition data further improves the degree of automation in data processing, while reducing dependence on the professional level of experimenters and reducing the complexity and error rate of manual operations.

[0028] More importantly, the accuracy and reliability of the data are ensured by introducing principal component analysis and noise-resistant Mahalanobis distance for outlier detection and removal. The dataset obtained after preprocessing and spectral wavelength selection provides strong support for building an accurate model.

[0029] Furthermore, the present invention provides an innovative hybrid intelligent learning model that integrates stacked autoencoders and random forest algorithms. This model can accurately predict the complex relationship between multiple batches of chemical composition data and density data and spectral data of poplars. Through this technological breakthrough, we have achieved high-throughput analysis of a large number of poplar samples, which not only significantly improves the detection efficiency, but also greatly reduces the detection cost, and reduces the dependence of experiments on chemicals, meeting environmental protection requirements. It realizes the quantitative prediction of the relationship between large batches of poplar wood properties and spectral data, greatly improves the adaptability and prediction accuracy of the model, ensures the reliability and stability of the results, simplifies the operation process, and reduces the professional requirements for experimenters.

[0030] Therefore, the high-throughput quantitative identification method of poplar wood properties provided by the present invention shows significant beneficial effects in high-throughput rapid analysis, cost reduction, environmental protection, data accuracy and model prediction accuracy, providing important technical support for forest breeding and resource optimization management, and effectively solving the problems existing in traditional wood property quantitative methods.

[0031] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0032] Specifically, the embodiment of the present invention provides a method for high-throughput quantitative identification of poplar wood properties, comprising: S1. Automatically identify and select poplar log segments that meet preset conditions, and control processing equipment to process the poplar log segments into samples.

[0033] Furthermore, if Figure 2 As shown, step S1 includes: S11. Utilize the sensor array to automatically identify and select poplar log segments that meet preset conditions.

[0034] In this step, a sensor array composed of high-resolution optical sensors, infrared sensors and ultrasonic sensors in a specific spatial layout is provided, and each sensor in the array is connected to the central processing unit through a flexible connecting line. Among them, the high-resolution optical sensor is used to capture the subtle texture changes on the surface of the wood and make preliminary identification of surface defects such as blue discoloration. Infrared sensor is used to penetrate the surface of the wood and detect possible defects inside, such as decay, cracks, etc. Ultrasonic sensor is used to analyze the density and structural changes inside the wood by emitting and receiving ultrasonic signals, and further confirm the type and location of defects.

[0035] Therefore, the sensor array can identify defects in the wood, such as cracks, knots, etc., and can also identify whether the wood has blue discoloration, that is, the color change of the wood due to moisture or fungal infection. At the same time, it can also distinguish the heartwood and sapwood of the wood. Based on the above identification process, the coordinates of the predetermined sampling area are obtained, and the required log sections can be screened and marked based on the coordinates of the predetermined sampling area.

[0036] S12. According to the recognition result, an instruction is sent to the processing equipment to control the processing equipment to process the selected poplar log segment into a blank sample of a predetermined size according to a preset processing flow.

[0037] After selecting the appropriate poplar log segments, instructions are automatically sent to the processing equipment. These instructions will accurately control the processing equipment to process the selected poplar log segments into blank samples of predetermined sizes and perform air drying according to the preset processing procedures, such as cutting and grinding. This automated processing method not only improves production efficiency, but also ensures the consistency of processing quality.

[0038] S13. After the rough sample is air-dried, the processing equipment is controlled to process the rough sample into a number of wood blocks, and the several small wood blocks are automatically divided into a first part of the sample and a second part of the sample, and each part of the sample is uniquely marked.

[0039] This step mainly processes the blank sample into several small wood blocks and automatically divides these small wood blocks into two parts: the first part of the sample and the second part of the sample. At the same time, each part of the sample is uniquely marked for accurate tracking and management in the subsequent process.

[0040] S14, controlling the pulverizer to pulverize the second part of the sample into wood powder, and controlling the filtering device to filter the wood powder into a sieve with a predetermined mesh number to obtain filtered wood powder, wherein the mesh number of the sieve ranges from 40 to 60 meshes.

[0041] S2. Automatically search and mark the area to be collected that meets the preset conditions on at least a part of the samples, directly collect the spectral data of the unmarked samples or collect the spectral data of the area to be collected of the marked samples through the spectrometer, and perform density testing and / or material property testing of multiple batches of samples.

[0042] Furthermore, if Figure 3 As shown, step S2 includes: S21. After the first part of the samples are placed in a constant temperature and humidity chamber for equilibrium treatment, the density detection device is controlled to detect the density standard value of the first part of the samples, and the density data is recorded as a reference label of the model.

[0043] In this step, the density of the poplar samples was measured by weighing method and used as the reference label for the subsequent model. In order to control the single factor, the poplar samples were placed in a constant temperature and humidity chamber for equilibrium treatment to achieve mass balance.

[0044] S22. Start and control the spectrometer to preheat for 25 to 35 minutes.

[0045] S23. After preheating is completed, the light source and instrument of the spectrometer are calibrated using a polytetrafluoroethylene reference white plate.

[0046] S24. Control the spectrometer in diffuse reflection mode to collect near infrared spectrum of the filtered wood powder in a constant temperature chamber.

[0047] S25. Use a camera deployed at the optical fiber probe of the spectrometer to read the surface information of the first part of the sample, and identify defects, blue discoloration, heartwood or sapwood again to search for the area to be collected; control the spectrometer to collect near-infrared spectrum data of the cross section of the first part of the sample in the area to be collected.

[0048] In one embodiment, the camera installed at the optical fiber probe of the spectrometer is first activated. The camera has high-definition shooting capabilities and can capture microscopic details of the wood surface. The camera continuously shoots the filtered wood powder and the surface of the first part of the sample in real time, capturing detailed image information including texture, color, defects (such as cracks, knots, blue discoloration) and material distinction (heartwood / sapwood). Once it is confirmed that the collection point is free of defects, blue discoloration and meets specific material requirements (such as giving priority to heartwood or sapwood), the data collection function of the near-infrared spectrometer is immediately started to accurately move the optical fiber probe to the sampling area to capture the near-infrared spectrum data of the point. If any situation that does not meet the preset conditions is found during the collection process (such as probe offset, new defects, etc.), the data collection will be immediately interrupted and repositioned to a new qualified sampling point.

[0049] In another embodiment, a Field Spec® near-infrared spectrometer (ASD, USA) is used in a constant temperature room at 24°C to collect near-infrared spectra of wood powder and a portion of the sample cross section. The wavelength range covers 350-2500nm, including the entire near-infrared spectrum band, and the recording wavelength interval is 1nm. When collecting near-infrared spectra, first turn on the machine for 30 minutes of preheating; then use a polytetrafluoroethylene reference white board to calibrate the light source and instrument, and perform a reference white board calibration every 30 minutes during the collection process; then, set the collection parameters so that each spectrum is averaged by 30 scans. In the experiment, the near-infrared spectrum data of wood was collected in diffuse reflectance mode.

[0050] S26. Performing material property tests on multiple batches of filtered wood powder according to preset standards, wherein the material properties of the multiple batches include the content of lignin, cellulose, and hemicellulose.

[0051] The "Determination of Structural Carbohydrates and Lignin in Biomass" method was determined as the preset standard for the material properties test. This is a standard method for analyzing the main components in biomass materials (such as wood, crop waste, etc.).

[0052] The specific tests are as follows: (1) Cellulose content determination: For each batch of wood powder samples, the cellulose content is determined using appropriate chemical reagents and instruments according to the method in the preset standard. The cellulose content data of each batch of wood powder is recorded and the accuracy and repeatability of the data are ensured.

[0053] (2) Hemicellulose content determination: For each batch of wood powder samples, continue to determine the hemicellulose content according to the method in the preset standard. Similarly, record the hemicellulose content data of each batch of wood powder and maintain the accuracy and consistency of the data.

[0054] (3) Determination of lignin content: For each batch of wood powder samples, the lignin content is determined according to the method specified in the preset standard. The lignin content data of each batch of wood powder is recorded to ensure the reliability and validity of the data.

[0055] S3, automatically import, match and store spectral data, density data and chemical composition data using pre-built automation scripts.

[0056] Furthermore, if Figure 4 As shown, step S3 includes: S31. Start the pre-configured Python script by timed start or event trigger, and use the incremental import mechanism to automatically detect and read the new result files generated by the spectrometer, material tester and density detection equipment.

[0057] Pre-build the Python script and configure its startup mode as follows: (1) Scheduled startup: Set a fixed time interval (such as every hour, every day, etc.), and automatically execute the Python script when the specified time arrives. This method is suitable for situations where the frequency of result file generation is relatively stable. (2) Event-triggered startup: Listen for result file generation events of spectrometers, material testers, and density detection equipment. Once a new file is generated, the execution of the Python script is triggered immediately. This method has higher real-time requirements and can ensure that the newly generated files are read in time. In addition, configure its input import mechanism as an incremental import mechanism to automatically detect and read new result files generated by spectrometers, material testers, and density detection equipment.

[0058] S32. Using the asynchronous coroutine framework of the Python script, an asynchronous task queue is created, and each new result file detected is added to the asynchronous task queue in parallel as an independent task. In the asynchronous loop, new result files are continuously taken out from the asynchronous task queue and data is imported from the new result files.

[0059] According to the project requirements and system resources, it is determined to use Python's asynchronous programming framework (such as asyncio). For asyncio, it is necessary to create an event loop, which is the core of asynchronous operations. It is responsible for scheduling and executing all asynchronous tasks and callbacks to achieve non-blocking I / O operations. Therefore, first, you need to create an event loop as the basis of the entire asynchronous task queue. Then use methods in the asyncio library such as ensure_future() or create_task() to add the imported function as a task to the event loop, and then use a suitable data structure such as asyncio.Queue as the task queue to set up a monitoring mechanism to detect new result files generated by spectrometers, material testers, and density detection equipment in real time. Once a new file is detected, a new data import task is created and added to the previously initialized asynchronous task queue. In the asynchronous loop, new result files are continuously taken out of the asynchronous task queue and data is imported from the new result files. Preferably, the queue can be checked regularly in the event loop, and the await keyword for asynchronous programming in Python can be used to asynchronously execute tasks in the queue. This ensures that high responsiveness and throughput can still be maintained when importing large amounts of data.

[0060] Through the above steps, an efficient and reliable asynchronous task queue system can be built to realize the automatic and parallel processing of result files generated by spectrometers, material testers and density detection equipment. This will greatly improve the efficiency and accuracy of data processing.

[0061] S33. Using the Python script, create and initialize an empty dictionary, configure the unique tag of the sample as the key of the dictionary, and initialize the value of the dictionary to an empty list for storing various imported data associated with the key.

[0062] S34, traverse the unique serial number of each newly detected sample, and check whether the unique serial number of each sample already exists in the dictionary.

[0063] S35. If the unique serial number does not exist in the dictionary, create a new serial number entry for the unique serial number, and add the imported spectral data, chemical composition data, and density data to the list corresponding to the new serial number entry in parallel; if the unique serial number already exists in the dictionary, choose to skip the serial number entry, append a new serial number entry, or update the list data corresponding to the serial number entry.

[0064] If the currently traversed unique serial number does not exist in the dictionary, the script will create a new entry for this serial number in the dictionary, that is, a new key-value pair. The key is the unique serial number of the current sample, and the value is an empty list. Then, the script will add the imported spectral data, chemical composition data, and density data to the newly created list corresponding to the unique serial number.

[0065] If the currently traversed unique serial number already exists in the dictionary, the script will decide whether to skip the serial number entry, append a new serial number entry (usually when the data needs to be distinguished by additional dimensions such as time or batch), or update the list data corresponding to the serial number entry (for example, overwrite the old data with the new data, or append the new data record to the list) based on the preset logic.

[0066] S36. Use Python scripts to scan and verify data integrity, and check whether the data under each list is complete through pre-defined integrity rules; wherein the pre-defined integrity rules include non-empty field check, data range verification, and format verification.

[0067] S37. For the detected incomplete data, add a pre-specified tag or move to a new temporary data table for storing the incomplete data.

[0068] S38. Use Python script to perform duplicate sample detection, automatically identify and mark duplicate data with unique serial numbers by adding another pre-specified label, and prompt the user to check or perform a merge operation.

[0069] S39. After verification is completed, the Python script is used to save the verified spectral data, density data and chemical composition data into the specified output file.

[0070] S4. Principal component analysis and anti-noise Mahalanobis distance are introduced to detect and remove outliers from the stored data, and then the data set is obtained after preprocessing and spectral wavelength selection.

[0071] Furthermore, if Figure 5 As shown, step S4 includes: S41. Construct a spectral data matrix according to the stored spectral data, and calculate the covariance matrix of the spectral data matrix.

[0072] S42. Perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues ​​and corresponding eigenvectors.

[0073] S43. According to the size of the eigenvalue, select the eigenvector that can explain the total variance to a predetermined proportion to form a transformation matrix.

[0074] S44. Projecting the spectral data matrix to a low-dimensional space formed by the selected eigenvectors according to the transformation matrix to obtain a score matrix after dimensionality reduction.

[0075] In one embodiment, first, using the spectral data matrix X Calculate the covariance matrix ∑ , to capture the correlation between the bands, the spectral data matrix can be expressed as: ; In the formula, X is the spectral data matrix, x ij Indicates i The sample in j The value of a spectral feature.

[0076] The covariance matrix of the spectral data matrix is: ; In the formula, ∑ is the covariance matrix, X is the spectral data matrix, X T For the matrix X The transpose of n is the number of samples.

[0077] Secondly, the covariance matrix ∑ Perform eigenvalue decomposition and obtain the eigenvalue And the corresponding eigenvector: .

[0078] Then, according to the size of the eigenvalue, select the eigenvector that can explain 95% of the total variance and form it into a transformation matrix E: ; Number of eigenvectors k Requirements: ; In the formula, p Spectral feature number; Finally, the data X is projected into a low-dimensional space composed of principal components, and the score matrix is ​​obtained as follows: ; In the formula, S is the score matrix after dimensionality reduction, X is the spectral data matrix, E is the transformation matrix composed of eigenvectors.

[0079] S45. In the principal component space, the noise-resistant Mahalanobis distance algorithm is introduced, and the mean and covariance matrix of the score matrix are adjusted by using the method based on the median and absolute median difference.

[0080] S46. For each sample, the Mahalanobis distance is calculated according to the mean and covariance matrix of the score matrix obtained based on the median and the absolute median difference.

[0081] In order to estimate the mean vector and covariance matrix of the score matrix S more robustly in the presence of outliers, a statistical method based on the median and median absolute deviation (MAD) is used to calculate the center position and dispersion to reduce the impact of outliers. Specifically, the mean vector is replaced by the median, and the estimation of the covariance matrix is ​​adjusted based on MAD to make the estimation more robust. For each sample i , Mahalanobis distance D i The calculation formula is: ; In the formula, D i It is a sample i The Mahalanobis distance, S i It is a sample i The score vector of and are the mean and covariance matrix of the score matrix adjusted based on the median and MAD, respectively. is the inverse of the covariance matrix.

[0082] S47, determining a basic threshold based on the mean and standard deviation of the Mahalanobis distance to determine the relative position of each spectral data in the principal component space. k Nearest neighbors, calculate each spectral data to the corresponding k The distances of the nearest neighbors are calculated and the average value is taken to obtain the local aggregation information; the basic threshold is dynamically adjusted according to the local aggregation information to obtain the current dynamic threshold.

[0083] The basic threshold is dynamically adjusted according to the local aggregation information. For samples with low local density (i.e. relatively isolated points), the threshold for anomaly detection can be appropriately lowered because these points are more likely to be anomalies. Conversely, for samples with high local density, the threshold can be appropriately increased.

[0084] Specifically, the basic thresholds are: ; In the formula, C is the dynamic threshold, mean is the mean Mahalanobis distance of all samples, std is the standard deviation of the Mahalanobis distance of all samples, md The suffix name.

[0085] The local aggregation information is: ; In the formula, ρ i is the spectral data sample point i The local aggregation information of KNN (i) is the spectral data sample point i of k The set of nearest neighbors, d ( i , j ) is the spectral data sample point i and j The distance between The dynamic thresholds are: ; In the formula, is the spectral data sample point i Dynamic anomaly detection threshold, is an adjustment factor greater than 0.

[0086] S48. If the Mahalanobis distance of the sample exceeds the current dynamic threshold, it is marked as an abnormal point, and all spectral data marked as abnormal are deleted from the output file storing the spectral data to obtain cleaned spectral data.

[0087] Specifically, if the sample i The Mahalanobis distanceD i If it exceeds the threshold, it will be marked as an outlier, all detected abnormal spectral data samples and the corresponding standard values ​​will be deleted from the data set, and the cleaned data will be saved in a new file for further analysis.

[0088] Furthermore, if Figure 6 As shown, it also includes: A41. Preprocess the cleaned spectral data by one or more of the following processes: noise removal, smoothing, baseline correction, standard variable transformation, and standard normal transformation.

[0089] In this embodiment, the standard normal variant (SNV) is used. This method is to standardize the spectral data to eliminate the influence of factors such as scattering effect and particle size. The data preprocessed by this method can be expressed as: ; in, is the original spectral data, is the spectral data after preprocessing.

[0090] A42. Calculate the information gain value between each wavelength in the preprocessed spectral data and the target variable one by one. Information gain is an indicator to measure the amount of information provided by a certain wavelength for predicting the target variable.

[0091] A43. Sort all wavelengths according to the calculated information gain value, and select the first γ wavelengths as characteristic wavelengths, where γ is the sorting threshold set according to demand or expert experience; use the similarity measure based on the spectral intensity value to calculate the correlation between the remaining wavelengths that are not selected as characteristic wavelengths and the target variable.

[0092] For the remaining wavelengths that are not selected as feature wavelengths, the correlation between them and the target variable is calculated using a similarity measurement method based on spectral intensity values ​​(such as cosine similarity, Pearson correlation coefficient, etc.).

[0093] A44. Sort the remaining wavelengths according to the magnitude of the correlation, and select the first λ wavelengths as the wavelengths that are viscous-correlated with the target variable.

[0094] The remaining wavelengths are sorted according to the magnitude of the correlation. Another sorting threshold λ is set, and the wavelengths ranked in the first λ are selected. These wavelengths are considered to have a "viscous correlation" with the target variable. The meaning of "viscous correlation" is that although some wavelengths do not directly show a strong correlation with the target variable through conventional indicators such as information gain, they may have an auxiliary role in the prediction or classification of the target variable in practical applications.

[0095] A45. Perform a redundancy check on the wavelengths that are sticky with the target variable to exclude wavelengths whose difference with the characteristic wavelength is less than or equal to the set difference threshold or are repeated with the characteristic wavelength.

[0096] A46. After redundancy check, the spectral data are reconstructed based on the remaining non-redundant wavelengths that are viscous related to the target variable and the selected characteristic wavelengths.

[0097] A47. Based on the reconstructed spectral data, density data and chemical composition data, the training set and test set are obtained.

[0098] Preferably, after the above processing, the wavelength selection generally selects data in the wavelength range of 350-2500nm, 780-2500nm, 900-1700nm, 1100-2500nm, and 450-2400nm for modeling and analysis. If the influence of noise is taken into account, the near-infrared spectrum in the 450~2400nm band is selected for modeling and analysis. Next, according to the reconstructed spectral data, density data, and chemical composition data, the training set and the test set are obtained by dividing them in a ratio of 2:1.

[0099] S5. According to the data set, a hybrid intelligent learning model based on stacked autoencoders and random forests is constructed and trained for the spectral data of poplar and the test data to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplar.

[0100] Furthermore, if Figure 7 As shown, step S5 includes: S51, obtaining a training set data file containing spectral data and chemical composition data labels from the data set, and dividing to obtain a first feature matrix of the spectral data X chem and the label vector of chemical composition data Y chem Specifically: ; X chem is a two-dimensional matrix, where rows represent samples and columns represent spectral features. Assume that m samples, each with n spectral features, then X chem The dimension is m × n . X chem Elements in x ij Indicates iThe sample in j The measured value of a spectral feature.

[0101] ; Y chem is a one-dimensional vector, each element of which represents the chemical composition data label of a sample. m samples, then Y chem The dimension is m. Y chem The element y in i Represents the chemical composition data label of the i-th sample, such as lignin content, cellulose content, etc.

[0102] S52, obtaining a training set data file containing spectral data and density data reference labels from the data set, and dividing to obtain a second feature matrix of the spectral data X density and the label vector of the density data Y density .

[0103] X density Data structures and X chem It is also a two-dimensional matrix of m×n, representing the measured values ​​of m samples on n spectral features. Mathematical expression (same as X chem same): ; Y density is a one-dimensional vector, each element of which represents the density data label of a sample. Y chem The same is m. Y density Elements in y i Indicates i Density data labels for samples.

[0104] ; S53. Determine and construct two parallel autoencoder paths of stacked autoencoders, each autoencoder path includes multiple network layers for extracting data features layer by layer.

[0105] S54, randomly initialize the weights and biases of each autoencoder path to ensure that the network has diversity at the beginning of training and can gradually adjust to the optimal state through learning. Then, the first feature matrix X chemand the second characteristic matrix X density The input is sent to two parallel autoencoder paths for independent training. Afterwards, the network parameters of the two autoencoder paths are optimized by minimizing the reconstruction error of each path. The formula for minimizing the reconstruction error is: ; In the formula, α is the weight parameter, N is the number of data points of the original data, is the original input data, is the reconstructed data output by the autoencoder, is the mean of the original data, is the mean of the reconstructed data, is the variance of the original data, is the variance of the reconstructed data, is the covariance between the original data and the reconstructed data, and c1 and c2 are constant terms introduced to avoid division by zero errors. This formula ensures that the two autoencoder paths can independently learn the effective representation of their respective feature matrices, and continuously optimize the network parameters by minimizing the reconstruction error, thereby improving the performance of the autoencoder.

[0106] S55. Apply the trained stacked autoencoder to the first feature matrix X chem and the second characteristic matrix X density , respectively extract the compressed first deep feature representation F chem And the compressed second deep feature representation F density . When the feature matrix passes through the stacked autoencoder, it goes through a series of encoding operations. Each network layer compresses and abstracts the data to extract higher-level features. This process is performed layer by layer, with each layer taking the output of the previous layer as input and producing a new, more compact feature representation. The final two deep feature representations capture the key information and potential structure in the original data, while reducing the dimensionality of the data, which is helpful for subsequent model training and prediction.

[0107] S56, using the extracted first feature representation F chem and the label vector of chemical composition data Y chem , train a random forest regression model for chemical composition prediction, and use the extracted second feature representation F density and the label vector of the density data y density, train a random forest regression model for density prediction.

[0108] S57. The two random forest regression models were evaluated using the cross-validation method, and the model performance was optimized by adjusting the model parameters.

[0109] The two random forest regression models are evaluated on the validation set using a cross-validation method (e.g., K-fold cross-validation). During the cross-validation process, the parameters of the random forest regression model (e.g., the number of decision trees, maximum depth, etc.) are also adjusted to find the optimal parameter combination. These parameter adjustments can be achieved through methods such as grid search, random search, or Bayesian optimization. By continuously trying different parameter combinations and evaluating the performance of the model, the model configuration that best suits the current dataset can be found.

[0110] After completing the full cross validation, the overall correlation coefficient is calculated to evaluate the prediction effect of the model on the complete data set. r for: ; In the formula, r is the overall correlation coefficient, For the i The actual value of the samples, Representative i The predicted value of samples, is the average of the actual values, is the average of the predicted values, N is the number of test set samples.

[0111] S58. For the new input poplar spectral data, feature extraction is first performed through the trained stacked autoencoder, and then the extracted new features are input into the optimized random forest regression model for prediction, and the quantitative prediction results of multiple batches of chemical composition data and density data are output respectively.

[0112] In a specific embodiment, poplar is used as a research object, and the density of poplar wood is accurately and quantitatively identified by near-infrared spectroscopy technology.

[0113] In this embodiment, the correlation coefficients of the poplar wood density data modeling are shown in Table 1.

[0114] Table 1 Correlation coefficients of poplar wood density data modeling

[0115] And, the prediction results of the poplar wood density prediction model in this embodiment on the test set are shown in Table 2.

[0116] Table 2 Prediction results of poplar wood density prediction model on the test set

[0117] In another specific embodiment, poplar is used as the research object, and the lignin, cellulose and hemicellulose components of poplar wood are accurately and quantitatively identified by near-infrared spectroscopy technology.

[0118] In this embodiment, the correlation coefficients of the modeling of the poplar wood property data are shown in Table 3.

[0119] Table 3 Correlation coefficients of poplar wood property data modeling

[0120]

[0121] In this embodiment, the prediction results of the poplar wood property prediction model on the test set are shown in Table 4. Figure 8 , we can know the comparison between the predicted value and the true value of lignin.

[0122] Table 4 Prediction results of the poplar wood property prediction model on the test set

[0123]

[0124] In addition, an embodiment of the present invention provides a high-throughput quantitative identification system for poplar wood properties, including: a sample control module, which is used to automatically identify and select poplar log segments that meet preset conditions, and control processing equipment to process poplar log segments into samples. A collection and testing module, which is used to collect spectral data of samples through a spectrometer, and perform density tests on a portion of the samples, and perform multiple batches of wood property tests on another portion of the samples. An automated processing module, which is used to automatically import, match, and store spectral data and chemical composition data using a pre-built automated script. A data set determination module, which is used to introduce principal component analysis and anti-noise Mahalanobis distance, detect and remove outliers on the stored data, and then obtain a data set after applying preprocessing and spectral wavelength selection. A result output module, which is used to construct and train a hybrid intelligent learning model based on stacked autoencoders and random forests for the spectral data and test data of poplars according to the data set, so as to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplars.

[0125] Furthermore, an embodiment of the present invention provides a high-throughput quantitative identification device for poplar wood properties, comprising: a multimodal database cluster, including at least one database, the multimodal database cluster supporting distributed storage and parallel processing; and an elastic cloud storage communicatively connected to the multimodal database cluster; wherein the elastic cloud storage stores instructions that can be executed by at least one database, and the instructions are executed by at least one database so that the at least one database can execute the high-throughput quantitative identification method for poplar wood properties as described above.

[0126] Specifically, the core component of the device is a multimodal database cluster. This cluster contains at least one database and is specially designed to support distributed storage and parallel processing functions. This means that the device can process a large amount of data at the same time, ensuring data processing efficiency and stability during high-throughput identification. In addition, the device is connected to an elastic cloud storage system. Elastic cloud storage not only provides massive storage space, but also stores instructions that can be executed by at least one database in the multimodal database cluster. When these instructions are executed by the database, they can drive the database to execute a high-throughput quantitative identification method for poplar wood properties. This design enables the device to flexibly call the required instructions during the identification process, further improving the intelligence and automation level of identification.

[0127] In general, the embodiment of the present invention integrates multimodal database cluster and elastic cloud storage technology to build an efficient and intelligent high-throughput quantitative identification device for poplar wood properties. This device can not only meet the needs of large-scale data processing, but also can realize accurate identification of poplar wood properties by executing specific instructions, providing strong support for research and application in related fields.

[0128] Furthermore, an embodiment of the present invention provides a computer-readable medium having computer-executable instructions stored thereon, and when the executable instructions are executed by a processor, the high-throughput quantitative identification method for poplar wood properties as described above is implemented.

[0129] The embodiment of the present invention also innovatively provides a computer-readable medium. This medium is not an ordinary storage tool, but is specially designed to store specific computer-executable instructions. When these instructions are read and executed by the processor, they can accurately guide the processor to perform a series of complex operations. The core of these operations is to realize the above-mentioned high-throughput quantitative identification method of poplar wood properties. This means that through this computer-readable medium, users can easily and quickly deploy and implement this advanced analysis method on any compatible computer or processing device. This computer-readable medium provides users with a convenient and efficient way to achieve high-throughput quantitative identification of poplar wood properties. It not only greatly simplifies the deployment of the analysis process, but also greatly improves the accuracy and efficiency of the analysis, bringing revolutionary convenience to poplar wood property research and related application fields.

[0130] In summary, the embodiments of the present invention provide a method, system, equipment and medium for high-throughput and accurate quantitative identification of poplar wood properties. By innovatively applying near-infrared spectroscopy technology, the present invention successfully realizes rapid and accurate quantitative identification of key components in poplar wood such as density, lignin, cellulose and hemicellulose, thereby demonstrating its significant advantages in many aspects. First, in terms of high-throughput analysis capability, the present invention can process a large number of poplar samples at one time, realize high-throughput analysis, not only significantly improve the detection efficiency, but also effectively reduce the detection cost. At the same time, the method reduces the dependence on chemicals and is more in line with the current strict requirements for environmental protection. Secondly, in terms of accuracy, the present invention accurately reveals the deep relationship between poplar wood properties and spectral data by cleverly combining a hybrid intelligent learning model based on stacked autoencoders and random forests, and realizes high-precision quantitative prediction. This innovative measure greatly enhances the adaptability and prediction accuracy of the model, and ensures the reliability and stability of the analysis results. In addition, compared with traditional detection methods, the present invention simplifies the operation process and reduces the requirements for the professional skills of experimental personnel. At the same time, under the premise of ensuring detection accuracy, it achieves high efficiency and low cost in parallel in poplar wood property testing, bringing substantial convenience to related industries.

[0131] In summary, the present invention has achieved remarkable breakthroughs in the three key areas of efficiency, accuracy and environmental protection of poplar wood property analysis, providing strong technical support for forest breeding and resource optimization management, and demonstrating broad application prospects and practical value.

[0132] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0133] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments after obtaining the basic inventive concepts.

[0134] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the technical solution of the present invention and its equivalent technology, the present invention should also include these modifications and variations.

Claims

1. A high-throughput, precise and quantitative identification method for poplar wood properties, characterized in that: include: Automatically identify and select poplar logs that meet preset conditions, and control processing equipment to process the poplar logs into samples; Automatically search and mark the area to be collected that meets the preset conditions on at least a part of the samples, directly collect the spectral data of the unmarked samples or collect the spectral data of the area to be collected of the marked samples through the spectrometer, and perform density test and / or material property test of multiple batches on the samples; Automatically import, match, and store spectral, density, and chemical composition data using pre-built automation scripts; Principal component analysis and anti-noise Mahalanobis distance are introduced to detect and remove outliers from the stored data, and then the data set is obtained after preprocessing and spectral wavelength selection. According to the data set, a hybrid intelligent learning model based on stacked autoencoders and random forests is constructed and trained for the spectral data of poplar and the test data to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplar.

2. The method for high-throughput, precise and quantitative identification of poplar wood properties according to claim 1, characterized in that: Automatically identify and select poplar logs that meet preset conditions, and control processing equipment to process poplar logs into samples including: Using a sensor array, poplar log segments that meet preset conditions are automatically identified and selected; According to the recognition result, an instruction is sent to the processing equipment to control the processing equipment to process the selected poplar log segment into a blank sample of a predetermined size according to a preset processing flow; After the blank sample is air-dried, the processing equipment is controlled to process the blank sample into a number of wood blocks, and the several small wood blocks are automatically divided into the first part of the sample and the second part of the sample, and each part of the sample is uniquely marked; The pulverizer is controlled to pulverize the second part of the sample into wood powder, and the filtering device is controlled to filter the wood powder into a sieve with a pre-specified mesh number to obtain filtered wood powder.

3. The method for high-throughput, precise and quantitative identification of poplar wood properties according to claim 2, characterized in that: Automatically search and mark the area to be collected that meets the preset conditions on at least a part of the samples, directly collect the spectral data of the unmarked samples or collect the spectral data of the area to be collected of the marked samples through the spectrometer, and perform density test and / or material property test of multiple batches of samples, including: After the first part of the samples are placed in a constant temperature and humidity chamber for equilibrium treatment, the density detection device is controlled to detect the density standard value of the first part of the samples, and the obtained density data is recorded as a reference label of the model; Start and control the spectrometer to preheat for 25 to 35 minutes. After the preheating is completed, calibrate the light source and instrument of the spectrometer using a polytetrafluoroethylene reference white board; The spectrometer was controlled in diffuse reflectance mode to collect near-infrared spectra of the filtered wood powder in a constant temperature chamber; The surface information of the first part of the sample is read by using a camera deployed at the optical fiber probe of the spectrometer, and the defects, blue discoloration, heartwood or sapwood are identified again to search for the area to be collected; Controlling the spectrometer to collect near infrared spectrum data of a first portion of the sample cross section in the area to be collected; The filtered wood powder is subjected to multiple batches of material property tests according to preset standards, wherein the material properties of the multiple batches include the content of lignin, cellulose and hemicellulose.

4. The method for high-throughput, precise and quantitative identification of poplar wood properties according to claim 2, characterized in that: Automatically import, match, and store spectral, density, and chemical composition data using pre-built automation scripts including: Start pre-configured Python scripts based on timed or event triggers, and use the incremental import mechanism to automatically detect and read new result files generated by spectrometers, material testers, and density testing equipment; Using the asynchronous coroutine framework of Python script, an asynchronous task queue is created, and each new result file detected is added to the asynchronous task queue in parallel as an independent task. In the asynchronous loop, new result files are continuously taken out from the asynchronous task queue and data is imported from the new result files. Use the Python script to create and initialize an empty dictionary, configure the unique tag of the sample as the key of the dictionary, and initialize the value of the dictionary to an empty list to store various imported data associated with the key; Iterate over the unique serial number of each newly detected sample and check whether the unique serial number of each sample already exists in the dictionary: If the unique serial number does not exist in the dictionary, a new serial number entry is created for the unique serial number, and the imported spectral data, chemical composition data, and density data are added to the list corresponding to the new serial number entry in parallel; If the unique serial number already exists in the dictionary, choose to skip the serial number entry, append a new serial number entry, or update the list data corresponding to the serial number entry; Use Python scripts to scan and verify data integrity, and check whether the data under each list is complete through pre-defined integrity rules; the pre-defined integrity rules include non-empty field check, data range verification, and format verification; For the detected incomplete data, add a pre-specified label or move to a new temporary data table for storing incomplete data; Use Python scripts to perform duplicate sample detection, automatically identify and mark duplicate data with unique serial numbers by adding another pre-specified tag, and prompt the user to check or perform a merge operation; After the verification is completed, the Python script is used to save the verified spectral data, density data, and chemical composition data into the specified output file.

5. The method for high-throughput, precise and quantitative identification of poplar wood properties according to claim 1, characterized in that: The introduction of principal component analysis and anti-noise Mahalanobis distance to detect and remove outliers in the stored data includes: According to the stored spectral data, a spectral data matrix is ​​constructed, and a covariance matrix of the spectral data matrix is ​​calculated; Perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues ​​and corresponding eigenvectors; According to the size of the eigenvalue, the eigenvector that can explain the total variance to a predetermined proportion is selected to form a transformation matrix; According to the transformation matrix, the spectral data matrix is ​​projected into a low-dimensional space composed of selected eigenvectors to obtain a score matrix after dimensionality reduction; In the principal component space, the noise-resistant Mahalanobis distance algorithm is introduced, and the median and absolute median difference based methods are used to adjust the mean and covariance matrix of the score matrix. For each spectral data, the Mahalanobis distance was calculated based on the mean and covariance matrix of the score matrix obtained based on the median and absolute median difference; Determine a basic threshold based on the mean and standard deviation of the Mahalanobis distance; Determine each spectral data in the principal component space k Nearest neighbors, calculate each spectral data to the corresponding k The distances of the nearest neighbors are calculated and averaged to obtain local aggregation information; Dynamically adjust the basic threshold according to the local aggregation information to obtain the current dynamic threshold; If the Mahalanobis distance of a sample exceeds the current dynamic threshold, it is marked as an abnormal point, and all spectral data marked as abnormal are deleted from the output file storing the spectral data to obtain the cleaned spectral data; in, The spectral data matrix is: ； In the formula, X is the spectral data matrix, x ij Indicates i The spectral data samples are j The covariance matrix of the spectral data matrix is: ； In the formula, ∑ is the covariance matrix, X is the spectral data matrix, X T is the transpose of matrix X, n is the number of samples; The characteristic values ​​are: ； Eigenvectors: ； Number of eigenvectors k The conditions are: ； In the formula, p Spectral feature number; The score matrix after dimensionality reduction is: ； In the formula, S is the score matrix after dimensionality reduction, X is the spectral data matrix, E is the transformation matrix composed of eigenvectors; The Mahalanobis distance is: ； In the formula, D i is a sample of spectral data i The Mahalanobis distance, S i is a sample of spectral data i The score vector of and are the mean and covariance matrices of the score matrix, respectively. is the inverse matrix of the covariance matrix; The basic threshold is: ； In the formula, C is the dynamic threshold, mean is the mean Mahalanobis distance of all samples, std is the standard deviation of the Mahalanobis distance of all samples; The local aggregation information is: ； In the formula, ρ i is the spectral data sample point i The local aggregation information of KNN (i) is the spectral data sample point i of k The set of nearest neighbors, d ( i , j ) is the spectral data sample point i and j The distance between The dynamic thresholds are: ； In the formula, is the spectral data sample point i Dynamic anomaly detection threshold, is an adjustment factor greater than 0.

6. The method for high-throughput, precise and quantitative identification of poplar wood properties according to claim 5, characterized in that: After applying preprocessing and spectral wavelength selection, the resulting data set includes: The cleaned spectral data is preprocessed by one or more of the following processes: noise removal, smoothing, baseline correction, standard variable transformation, and standard normal transformation; Calculate the information gain value between each wavelength in the preprocessed spectral data and the target variable one by one; According to the calculated information gain value, all wavelengths are sorted, and the wavelengths ranked first γ are selected as characteristic wavelengths, where γ is a sorting threshold set according to demand or expert experience; Using a similarity measure based on spectral intensity values, the correlation between the remaining wavelengths that were not selected as feature wavelengths and the target variable is calculated; According to the size of the correlation, sort the remaining wavelengths and filter them to the top. λ The wavelength of each is regarded as the wavelength that is stickyly related to the target variable; Perform redundancy check on wavelengths that are stickyly related to the target variable to exclude wavelengths whose difference with the characteristic wavelength is less than or equal to the set difference threshold or are repeated with the characteristic wavelength; After the redundancy check, the spectral data are reconstructed based on the remaining non-redundant wavelengths that are stickyly related to the target variable and the selected characteristic wavelengths; A data set is obtained based on the reconstructed spectral data, density data and chemical composition data.

7. The method for high-throughput, precise and quantitative identification of poplar wood properties according to claim 1, characterized in that: According to the data set, a hybrid intelligent learning model based on stacked autoencoders and random forests is constructed and trained for the spectral data of poplar and the test data to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplar, including: Obtain training set data containing spectral data and chemical composition data labels from the data set, and divide it to obtain the first feature matrix of the spectral data X chem and the label vector of chemical composition data Y chem ; Obtain training set data containing reference labels of spectral data and density data from the data set, and divide it to obtain the second feature matrix of spectral data X density and the label vector of the density data Y density ; Determine and construct two parallel autoencoder paths of stacked autoencoders, each autoencoder path includes multiple network layers for extracting features of data layer by layer; Randomly initialize the weights and biases of each autoencoder path; The first characteristic matrix X chem and the second characteristic matrix X density The data is fed into two parallel autoencoder paths for independent training, and the network parameters of the two autoencoder paths are optimized by minimizing the reconstruction error of each path. Apply the trained stacked autoencoder to the first feature matrix X chem and the second characteristic matrix X density , respectively extract the compressed first deep feature representation F chem And the compressed second deep feature representation F density ; Using the extracted first feature representation F chem and the label vector of chemical composition data Y chem , train a random forest regression model for chemical composition prediction, and use the extracted second feature representation F density and the label vector of the density data y density , train a random forest regression model for density prediction; The two random forest regression models were evaluated using cross-validation methods, and the model performance was optimized by adjusting the model parameters; For the new input poplar spectral data, the trained stacked autoencoder is first used to extract features, and then the extracted new features are input into the optimized random forest regression model for prediction, and the quantitative prediction results of multiple batches of chemical composition data and density data are output respectively; The formula for minimizing the reconstruction error is: ； In the formula, α is the weight parameter, N is the number of data points of the original data, is the original input data, is the reconstructed data output by the autoencoder, is the mean of the original data, is the mean of the reconstructed data, is the variance of the original data, is the variance of the reconstructed data, is the covariance between the original and reconstructed data, c 1 and c 2 is a constant term introduced to avoid division by zero errors.

8. A high-throughput, precise and quantitative identification system for poplar wood properties, which is applied to implement the high-throughput, precise and quantitative identification method for poplar wood properties as described in any one of claims 1 to 7, characterized in that: include: A sample control module is used to automatically identify and select poplar log segments that meet preset conditions, and control processing equipment to process the poplar log segments into samples; The acquisition test module is used to acquire the spectral data of the samples through the spectrometer, and to perform density test on a part of the samples, and to perform material property test on another part of the samples in multiple batches; An automated processing module for automatically importing, matching, and storing spectral and chemical composition data using pre-built automation scripts; The data set determination module is used to introduce principal component analysis and anti-noise Mahalanobis distance to detect and remove abnormal points in the stored data, and then obtain the data set after applying preprocessing and spectral wavelength selection; The result output module is used to construct and train a hybrid intelligent learning model based on stacked autoencoders and random forests for the spectral data of poplar and the test data according to the data set, so as to output quantitative prediction results of multiple batches of chemical composition data and / or density data of poplar.

9. A high-throughput, precise and quantitative identification device for poplar wood properties, characterized in that: include: A multimodal database cluster, comprising at least one database, wherein the multimodal database cluster supports distributed storage and parallel processing; and elastic cloud storage that is communicatively connected to the multimodal database cluster; The elastic cloud storage stores instructions that can be executed by at least one database, and the instructions are executed by at least one database so that the at least one database can execute the high-throughput quantitative identification method for poplar wood properties as described in any one of claims 1-7.

10. A computer-readable medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by the processor, the high-throughput and accurate quantitative identification method of poplar wood properties as described in any one of claims 1 to 7 is implemented.

Images